Insights

The waiting room before the waiting room

Stephany Venuto, RDH & Shalin R. Shah, DMD, MS · August 24, 2026 · 13 min read

There is a stretch of every patient's story that no one in the practice ever sees. It begins the moment they decide that whatever is going on is worth doing something about, and it ends when they finally sit down across from you. Sometimes that stretch is an afternoon. Often it is three weeks. Everything that happens inside it — the call that rang out, the hold music, the portal password they could not remember, the voicemail nobody returned — is happening to someone who has already chosen you. Lately there is a loud argument about who ought to be answering during that stretch, and the people on both sides are not really arguing about software. One side is protecting the patient who cannot get through. The other is protecting something real that happens when a person picks up. We want to lay both out with the actual numbers, including the ones that do not flatter the side we started on. One disclosure first, since we are about to ask you to check everybody else's motives: we do not sell AI phone systems, and there is no version of your answer that pays us.

Almost nothing patients judge you on happens in the treatment room

Ask the people who actually run practices what they most want to fix, and they do not say diagnosis. At the end of last year MGMA put that question to medical group leaders for the year ahead, and the answers landed in a revealing cluster: no-shows first at 27%, online scheduling next at 24%, then phone access at 22% and wait times at 21%. Four problems, nearly evenly weighted, and not one of them is clinical. Every single one of them lives in the gap — before the visit, or in the hole where a visit was supposed to be. The care itself is not what is keeping these leaders up at night. It is the getting there.

It is worth picturing that gap from the other side, because from inside a practice it looks like a metric and from outside it looks like a Tuesday. Someone is calling on a lunch break, with a kid on one hip, on the third try. They get hold music. They wait ninety seconds, which is a great deal longer than it sounds when you are standing in a parking lot, and then they hang up, because the break is twenty minutes and this is not the only thing on it. Nothing has gone wrong, exactly. Nobody was rude. Nobody made a mistake. A person simply did not get through — and the story most people tell themselves next is not "the office is busy." It is "I'll deal with it later." Later is a word that does a great deal of quiet damage in healthcare, and it never shows up on a report, because the patient it happened to is not in your system to be counted.

The people absorbing all of this are the ones most likely to leave

There is one desk where the entire gap lands, and we do not talk about it enough. Every call that finally gets answered arrives carrying whatever the caller has been feeling while they waited — every coverage question asked in a voice working hard to stay steady, every person who is frightened and expressing it as impatience. That is the front desk's day, layered on top of the actual work of the front desk. MGMA's staff turnover poll in May of this year found the picture roughly holding: of 303 groups, 39% said turnover was about the same as last year and 30% said it was lower, while 28% said it was higher. Stabilized, not solved. And among the groups where it had gotten worse, the roles named most often were medical assistants and front-office staff — the exact seats where patient access is either protected or quietly lost.

That turnover has a sound you can hear on the phone. When the desk is short, calls get abandoned, scheduling gets less accurate, complaints go up — which makes the next shift harder, which makes the seat harder to keep filled. It is a loop, and it does not open by asking people to care more. They already care more; that is precisely why the seat is hard. When the AMA surveyed nearly 1,700 physicians this year, seven in ten said they saw AI as a way to automate the tasks that feed burnout. That is the pressure this whole argument is being had under, and it is worth naming before anybody takes a side.

The honest case for letting a machine pick up

Start with the strongest thing this side has going for it, which is not a product demo. It is that the real alternative to a machine answering is usually nothing answering. In Invoca's healthcare report this May, 83% of people in the middle of a significant healthcare decision said they had already hung up on somebody because the hold went on too long, and 85% said they would simply move to whoever got back to them faster. About a third said they reach for the automated option deliberately — not because they enjoy it, but because they have done the arithmetic on hold music and decided the machine is the shorter road. Sit with that for a moment, because it reframes the argument entirely. A tool that answers on the first ring at 7:40 on a Thursday evening is not competing with your best team member on her best day. It is competing with a ring tone, and a ring tone has never once helped anybody.

The formal evidence is thinner than the marketing, though it is not nothing. At the international AI-in-medicine conference in Ottawa this July, a peer-reviewed evaluation reported that voice agents contacting roughly 8,800 members of a virtual gastrointestinal practice over seven weeks lifted completed appointments about three and a half times over and saved something like 75 hours of staff time per thousand contacts, with no recorded safety incidents. Then read the fine print: the study was run by the company that builds the agents, the practice was virtual-first rather than a clinic with a lobby, the calls went out rather than coming in, and fewer than a quarter were answered at all. A real result in a narrow setting, not a promise about your front desk. And then there is the finding that makes everyone uncomfortable, ourselves included. When licensed clinicians blindly rated answers to real patient questions from a public forum for JAMA Internal Medicine, they judged the chatbot's replies empathetic or very empathetic 45.1% of the time and the physicians' 4.6% of the time. The physicians were volunteering fast answers between patients and the machine had nothing else to do that day, so it is hardly a fair fight — but it closes off one line of argument for good. The strongest case for people is not that a machine cannot sound warm. It is that patients do not want warmth from a machine.

And they say so, more clearly than the vendors let on

Whenever anybody asks patients directly, the answer comes back the same way. In that June survey of 6,000 adults, 90% said they would rather speak to a real person than to AI when contacting a healthcare practice, 71% said human agents show more genuine empathy and care, and 55% would not trust an automated service to take the right action or pass along accurate information. Two-thirds were uneasy about an AI system holding that much of their personal data at all. Apply the same scrutiny there we asked you to apply to the vendors: that survey was paid for by a company selling human receptionists, and it is not a neutral instrument. So look at the number from the other camp's own research, where a call-analytics firm with every commercial reason to be bullish on automation still found 60% of healthcare consumers reaching for a person when both options were equally available, and 87% saying human connection matters when the decision is a serious one. The two disagree about nearly everything except direction. And one more finding, which ought to interest anyone who has wondered how they compare to the practice down the road: 78% said that between two similar options, they would choose the one where a human being answered the phone.

There is money underneath this too, and it has been sitting there for years. Deloitte's analysis of patient-experience scores against hospital finances from 2008 through 2014 found that hospitals rated excellent by their patients ran an average net margin of 4.7%, against 1.8% for the ones rated low, and that a ten-point gain in top-box ratings tracked with roughly a 1.4% gain in margin. Those are hospitals rather than practices, that is correlation rather than cause, and the data is old enough now to be handled gently. But the part worth carrying forward is what correlated most strongly, which was not the building and not the technology. It was nurse-patient engagement — one person, paying attention, to another person. If the warmth of your team were only a lovely thing, this would be a values conversation and we would happily have it on those terms. It is also on the ledger. One last number, because it governs how you ought to behave whichever way you land: 57% of people say they cannot reliably tell whether they are speaking to a person or a machine, and 85% say the machine ought to tell them which it is. Trust does not break when a tool answers the phone. It breaks when somebody works out afterward that it did, and we wrote at length about saying it out loud — every word of which applies to the voice on your phone line.

What a machine should be allowed to say when it picks up

If you come down on the side of letting something answer, be precise about which job you are handing over, because the front desk is really two jobs wearing one badge. The first is logistics — is there an opening, what should I bring, do you take this plan, can I move Thursday. That work is repetitive, time-sensitive, arrives in floods at hours nobody is staffed for, and a machine does it well and without resentment. The second job is judgment: hearing that the voice on the line is more worried than the words are, and understanding that this is not actually a scheduling call. That one is not automatable, and anyone selling it to you as though it were is selling you a risk with a friendly interface. So draw the line where it belongs. The tool answers, says plainly that it is a tool without being asked, gathers what is needed, and books only against real availability. It does not assess symptoms, it does not decide urgency, it does not improvise about coverage or cost, and it never talks anybody out of being seen. And it hands off — fast, to a named human, with everything already gathered so the patient never repeats themselves — the moment someone asks for a person or says something the system was not built for. The escalation path is not a feature of the design. It is the design.

Then measure the things a patient can feel rather than the things that flatter the purchase. How long until anyone or anything answers. How many callers hang up before they reach anybody at all. Of the people who ask for a human, how many get one, and how quickly. How many callbacks promised by a specific time actually happened by that time. You will also run into impressive recovered-revenue figures published by the companies selling these systems, and those deserve the same scrutiny as any number with a sales motive behind it — which practices, over how long, measured by whom, compared against what. Your own thirty days of phone data will tell you more than any of it, and you already own that.

Staying reachable without handing anyone the phone

Plenty of good practices will land on the other side of this, for reasons that have nothing to do with being behind, and if that is you then the work is not to defend the choice. It is to make the choice real — because a patient standing in a parking lot cannot hear the difference between a practice that believes in human connection and a practice that simply did not pick up. Start where the practice above started, with your own call log; the misses are rarely spread evenly, and moving one break covers more ground than most purchases would. After that, the strongest move available to you is letting people book without a conversation at all. Online self-scheduling takes an entire category of routine calls off the desk without putting a synthetic voice anywhere near a patient, and it is no accident practice leaders ranked it second in that MGMA poll. Then give people a way to reach a human that is not the phone: a texting line a named person actually watches, with a stated window for replies. Replace hold with a promise — a callback at a specific hour, offered out loud and kept — because a kept callback builds more trust than a shorter queue ever will. For nights and weekends, a live answering service staffed by actual people is still a real thing you can buy, and so is a shared after-hours rotation with a practice you trust down the road. And if it truly comes down to voicemail, let the greeting name a time — every message returned by noon the next business day — and then be the practice that does it. None of this is old-fashioned. It is a set of promises, and the only thing that ever makes promises fail is nobody owning them.

Three questions that decide which company you pick

If you do go looking, the choosing matters more than the choice. Vendors are converging fast on the same feature list, and the ones that will hurt you look nearly identical on a website to the ones that will help. Three questions do most of the sorting, and we would ask them in this order. First, make them show you the handoff in numbers: what share of callers who ask for a person actually get one, how many seconds that takes, and what happens at six in the evening when there is no person to hand to. If the answer arrives as a demo rather than a number, you have already learned the thing you needed to know. Second, settle the data terms in writing before anything else — a signed business associate agreement, where recordings and transcripts live, how long they are kept, and specifically whether your patients' voices are used to train anybody's models and exactly how you decline. Third, insist that it books only against real availability in your actual schedule and writes back into your actual system — and get the refusal list from the section above signed rather than described, so that what the tool will not do is a term of the contract instead of a promise made on a call.

After those three, the rest of the list goes quickly. It should identify itself as a tool, unmistakably and without being asked. Ask for a reference practice of your size and your specialty, and then actually call their main line at a bad hour, from your own phone, the way a patient would — that ten minutes will tell you more than the sales call did. Take the shortest contract you can get, because the first ninety days will teach you things no demo can. Name one person inside the practice who owns the thing and give them the authority to switch it off without calling a meeting. And be wary in direct proportion to how confident the pitch is. In our experience the companies worth trusting are the ones who volunteer what their system is bad at before you get around to asking.

So the argument was never really about technology. Being reachable is not a business function that happens to sit near care. For the person on the other end of the line, it is the first act of care they receive from you, and it happens before you know their name. Presence — the thing that makes a patient feel genuinely held — is not only what you offer once someone is sitting in front of you. It is also whether they could find you at all on the day they finally worked up to asking for help. There is an honest version of this with a machine on the line and an honest version without one, and the practices that get it wrong are almost never the ones who chose differently than you would have. They are the ones who never really chose. Most practices are already good at presence once the door closes and the visit begins. The work worth doing now is in everything that happens before it opens.